NYC Traffic Accidents - Exploratory Data Analysis (Python)

import pandas as pd
import numpy as np
import plotly.graph_objects as go
import plotly.express as px
import plotly.io as pio

# pio.renderers.default = "notebook_connected"

clean_path = "../data/nyc_traffic_accidents_clean.csv"
df = pd.read_csv(clean_path)
df.shape, df.dtypes.head()

((74881, 28),
 crash_date     object
 borough        object
 zip_code      float64
 latitude      float64
 longitude     float64
 dtype: object)

1. Monthly Trends

monthly = (
    df.groupby(["month", "month_name"])
      .agg(
          total_crashes=("collision_id", "count"),
          fatal_crashes=("is_fatal", "sum")
      )
      .reset_index()
      .sort_values("month")
)
total_fatal_crashes = sum(monthly["fatal_crashes"])
monthly["fatal_rate_pct"] = round((monthly["fatal_crashes"] / total_fatal_crashes) * 100, 2)
monthly

	month	month_name	total_crashes	fatal_crashes	fatal_rate_pct
0	1	January	14287	17	12.59
1	2	February	13684	20	14.81
2	3	March	11057	8	5.93
3	4	April	4116	13	9.63
4	5	May	6149	12	8.89
5	6	June	7616	30	22.22
6	7	July	9225	20	14.81
7	8	August	8747	15	11.11

fig = go.Figure()

# Bars: total crashes
fig.add_trace(go.Bar(
    x=monthly["month_name"],
    y=monthly["total_crashes"],
    name="Total Crashes"
))

# Line: fatality rate
fig.add_trace(go.Scatter(
    x=monthly["month_name"],
    y=monthly["fatal_rate_pct"],
    name="Share of Fatal Crashes (%)",
    mode="lines+markers",
    yaxis="y2"
))

fig.update_layout(
    title="Monthly Collision Trends (Jan-Aug 2020)",
    xaxis_title="Month",
    yaxis_title="Number of Crashes",
    yaxis2=dict(
        title="Share (%)",
        overlaying="y",
        side="right"
    ),
    barmode="group",
    template="plotly_white"
)

fig.show()

2. Day-of-Week Analysis

weekday_order = [
    "Sunday", "Monday", "Tuesday", "Wednesday", "Thursday", "Friday", "Saturday"
]

df["day_of_week"] = pd.Categorical(
    df["day_of_week"],
    categories=weekday_order,
    ordered=True
)

dow = (
    df.groupby("day_of_week", observed=True)
      .agg(
          total_crashes=("collision_id", "count"),
          fatal_crashes=("is_fatal", "sum")
      )
      .reset_index()
      .sort_values("day_of_week")
)

dow["fatal_rate_pct"] = round((dow["fatal_crashes"] / total_fatal_crashes) * 100, 2)
dow

	day_of_week	total_crashes	fatal_crashes	fatal_rate_pct
0	Sunday	9003	22	16.30
1	Monday	10511	16	11.85
2	Tuesday	10613	25	18.52
3	Wednesday	10638	15	11.11
4	Thursday	11244	13	9.63
5	Friday	12271	16	11.85
6	Saturday	10601	28	20.74

fig = go.Figure()

# Bars: total crashes
fig.add_trace(go.Bar(
    x=dow["day_of_week"],
    y=dow["total_crashes"],
    name="Total Crashes"
))

# Line: fatality rate
fig.add_trace(go.Scatter(
    x=dow["day_of_week"],
    y=dow["fatal_rate_pct"],
    name="Share of Fatal Crashes (%)",
    mode="lines+markers",
    yaxis="y2"
))

fig.update_layout(
    title="Crashes by Day of Week",
    xaxis_title="Day of Week",
    yaxis_title="Number of Crashes",
    yaxis2=dict(
        title="Share (%)",
        overlaying="y",
        side="right"
    ),
    template="plotly_white"
)

fig.show()

3. Hour-of-Day Patterns

hourly = (
    df.groupby("hour")
      .agg(
          total_crashes=("collision_id", "count"),
          fatal_crashes=("is_fatal", "sum")
      )
      .reset_index()
      .sort_values("hour")
)

hourly.head()

	hour	total_crashes	fatal_crashes
0	0	2948	7
1	1	1474	7
2	2	1139	2
3	3	989	7
4	4	975	7

fig = go.Figure()

fig.add_trace(go.Scatter(
    x=hourly["hour"],
    y=hourly["total_crashes"],
    name="Total Crashes",
    mode="lines+markers"
))

fig.update_layout(
    title="Crashes by Hour of Day",
    xaxis=dict(
        title="Hour of Day",
        dtick=1,
        tickvals=hourly["hour"]
    ),
    yaxis_title="Total Crashes",
    template="plotly_white"
)

fig.show()

weekday_df = df[~df["is_weekend"]]
weekend_df = df[df["is_weekend"]]

weekday_hourly = weekday_df.groupby("hour")["collision_id"].count() / weekday_df["crash_date"].nunique()
weekend_hourly = weekend_df.groupby("hour")["collision_id"].count() / weekend_df["crash_date"].nunique()

hourly_avg = pd.DataFrame({
    "hour": range(24),
    "weekday_avg": weekday_hourly.reindex(range(24), fill_value=0).values,
    "weekend_avg": weekend_hourly.reindex(range(24), fill_value=0).values
})

hourly_avg.head()

	hour	weekday_avg	weekend_avg
0	0	11.052023	15.014493
1	1	4.809249	9.304348
2	2	3.520231	7.681159
3	3	2.907514	7.043478
4	4	3.069364	6.434783

fig = go.Figure()

fig.add_trace(go.Scatter(
    x=hourly_avg["hour"],
    y=hourly_avg["weekday_avg"],
    name="Weekday Avg",
    mode="lines+markers"
))

fig.add_trace(go.Scatter(
    x=hourly_avg["hour"],
    y=hourly_avg["weekend_avg"],
    name="Weekend Avg",
    mode="lines+markers"
))

fig.update_layout(
    title="Average Crashes by Hour",
    xaxis=dict(
        title="Hour of Day",
        dtick=1,
        tickvals=hourly_avg["hour"]
    ),
    yaxis_title="Avg Crashes per Day",
    template="plotly_white"
)

fig.show()

radar_df = hourly_avg.copy()
radar_df.loc[24] = radar_df.loc[0] # close the loop
radar_df["hour_label"] = radar_df["hour"].astype(int).astype(str)

fig = go.Figure()

fig.add_trace(go.Scatterpolar(
    r=radar_df["weekday_avg"],
    theta=radar_df["hour_label"],
    name="Weekday Avg",
))

fig.add_trace(go.Scatterpolar(
    r=radar_df["weekend_avg"],
    theta=radar_df["hour_label"],
    name="Weekend Avg",
))

fig.update_layout(
    title="Radar Chart - Weekday vs Weekend Hourly Patterns",
    template="plotly_white",
    polar=dict(
        radialaxis=dict(visible=True),
        angularaxis=dict(direction="clockwise")
    )
)

fig.show()

4. Day × Hour Heatmap

day_hour = (
    df.groupby(["day_of_week", "hour"], observed=True)["collision_id"]
      .count()
      .reset_index()
      .pivot(index="day_of_week", columns="hour", values="collision_id")
      .reindex(index=weekday_order)  # just to be explicit
)

all_hours = list(range(24))
day_hour = day_hour.reindex(columns=all_hours, fill_value=0)

day_hour

hour	0	1	2	3	4	5	6	7	8	9	...	14	15	16	17	18	19	20	21	22	23
day_of_week
Sunday	510	312	271	254	232	195	182	165	231	241	...	615	537	566	539	526	462	407	356	365	292
Monday	385	200	126	112	117	142	310	407	591	552	...	682	690	746	736	658	494	373	325	278	219
Tuesday	333	134	88	96	78	159	294	433	643	576	...	719	654	799	750	693	509	444	326	339	264
Wednesday	367	154	129	106	103	176	258	407	563	509	...	741	730	756	743	688	520	402	378	366	259
Thursday	391	144	116	78	118	148	316	445	679	593	...	722	683	814	778	703	556	454	399	340	283
Friday	436	200	150	111	115	178	315	383	648	638	...	815	781	870	804	800	628	574	472	430	428
Saturday	526	330	259	232	212	180	193	223	323	330	...	722	602	668	624	628	569	484	455	439	419

7 rows × 24 columns

fig = px.imshow(
    day_hour,
    labels=dict(
        x="Hour of Day",
        y="Day of Week",
        color="Number of Crashes"
    ),
    x=day_hour.columns,
    y=day_hour.index,
    aspect="auto",
    color_continuous_scale="Reds"
)

fig.update_layout(
    title="Crashes Heatmap - Day of Week × Hour of Day",
    xaxis=dict(
        tickvals=day_hour.columns
    ),
    template="plotly_white"
)

fig.show()

5. High-Risk Locations

streets = df[df["on_street_name"].notna()].copy()
streets["on_street_name"] = streets["on_street_name"].str.strip()

top_streets = (
    streets.groupby("on_street_name")
        .agg(total_crashes=("collision_id", "count"))
        .reset_index()
        .sort_values("total_crashes", ascending=False)
        .head(10)
)

# Add % of total collisions
total_crashes_citywide = len(df)
top_streets["pct_of_total"] = round((top_streets["total_crashes"] / total_crashes_citywide) * 100, 2)

top_streets

	on_street_name	total_crashes	pct_of_total
942	BELT PARKWAY	1241	1.66
2403	LONG ISLAND EXPRESSWAY	745	0.99
1051	BROOKLYN QUEENS EXPRESSWAY	738	0.99
1834	FDR DRIVE	728	0.97
2455	MAJOR DEEGAN EXPRESSWAY	591	0.79
1976	GRAND CENTRAL PKWY	581	0.78
1031	BROADWAY	575	0.77
742	ATLANTIC AVENUE	532	0.71
1322	CROSS BRONX EXPY	526	0.70
1325	CROSS ISLAND PARKWAY	512	0.68

fig = go.Figure()

fig.add_trace(go.Bar(
    x=top_streets["pct_of_total"],
    y=top_streets["on_street_name"],
    orientation="h",
    name="Share of Total Collisions (%)",
    text=top_streets["pct_of_total"].astype(str) + "%",
    textposition="outside"
))

fig.update_layout(
    title="Top 10 Streets by Collision Frequency",
    xaxis_title="Share of Total Collisions (%)",
    yaxis_title="Street",
    template="plotly_white",
)

fig.show()

6. Contributing Factors

cf = df.copy()

# normalize strings
cf["contributing_factor_vehicle_1"] = (
    cf["contributing_factor_vehicle_1"]
      .fillna("Unspecified")
      .str.strip()
)

# eliminate weird placeholders (Excel often had blanks)
cf["contributing_factor_vehicle_1"] = (
    cf["contributing_factor_vehicle_1"]
        .replace("", "Unspecified")
        .fillna("Unspecified")
)
cf_clean = cf[cf["contributing_factor_vehicle_1"] != "Unspecified"]

cf_all = (
    cf_clean.groupby("contributing_factor_vehicle_1")
      .agg(total_crashes=("collision_id", "count"))
      .reset_index()
      .sort_values("total_crashes", ascending=False)
      .head(10)
)

total_crashes_citywide = len(df)
cf_all["pct_of_total"] = ((cf_all["total_crashes"] / total_crashes_citywide) * 100).astype(int)

cf_all

	contributing_factor_vehicle_1	total_crashes	pct_of_total
8	Driver Inattention/Distraction	19123	25
17	Following Too Closely	5202	6
14	Failure to Yield Right-of-Way	4815	6
32	Passing or Lane Usage Improper	2840	3
4	Backing Unsafely	2829	3
31	Passing Too Closely	2687	3
27	Other Vehicular	2233	2
49	Unsafe Speed	2015	2
48	Unsafe Lane Changing	1809	2
46	Traffic Control Disregarded	1544	2

fig = px.treemap(
    cf_all,
    path=["contributing_factor_vehicle_1"],
    values="pct_of_total",
    color="pct_of_total",
    color_continuous_scale="Blues",
    title="Top 10 Contributing Factors - Share of Total Collisions (%)"
)
fig.data[0].texttemplate="%{label}<br>%{value}%"
fig.show()

cf_fatal = (
    cf_clean[cf_clean["is_fatal"]]
      .groupby("contributing_factor_vehicle_1")
      .agg(fatal_crashes=("collision_id", "count"))
      .reset_index()
      .sort_values("fatal_crashes", ascending=False)
)

total_fatal = cf_fatal["fatal_crashes"].sum()
cf_fatal["pct_of_fatal"] = ((cf_fatal["fatal_crashes"] / total_fatal) * 100).astype(int)

cf_fatal.head(10)

	contributing_factor_vehicle_1	fatal_crashes	pct_of_fatal
16	Unsafe Speed	33	32
14	Traffic Control Disregarded	15	14
3	Driver Inattention/Distraction	13	12
6	Failure to Yield Right-of-Way	10	9
13	Pedestrian/Bicyclist/Other Pedestrian Error/Co...	6	5
9	Illnes	5	4
4	Driver Inexperience	4	3
17	View Obstructed/Limited	3	2
2	Backing Unsafely	3	2
1	Alcohol Involvement	2	1

fig = px.treemap(
    cf_fatal.head(5),
    path=["contributing_factor_vehicle_1"],
    values="pct_of_fatal",
    color="pct_of_fatal",
    color_continuous_scale="Reds",
    title="Top Contributing Factors - Fatal Collisions (%)"
)
fig.data[0].texttemplate="%{label}<br>%{value}%"
fig.show()

7. Vehicle Types & Victims

vt = df.copy()
vt["vehicle_type_code_1"] = vt["vehicle_type_code_1"].fillna("Unspecified").str.strip()
vt_all = (
    vt.groupby("vehicle_type_code_1")
      .agg(total_crashes=("collision_id", "count"))
      .reset_index()
      .sort_values("total_crashes", ascending=False)
)

total_crashes = len(df)
vt_all["pct_of_total"] = round(vt_all["total_crashes"] / total_crashes * 100, 1)

vt_all_clean = vt_all[vt_all["vehicle_type_code_1"] != "Unspecified"]

vt_all_clean.head(10)

	vehicle_type_code_1	total_crashes	pct_of_total
179	Sedan	34349	45.9
182	Station Wagon/Sport Utility Vehicle	27541	36.8
196	Taxi	2768	3.7
162	Pick-up Truck	1882	2.5
19	Box Truck	1417	1.9
22	Bus	950	1.3
17	Bike	882	1.2
203	Tractor Truck Diesel	587	0.8
138	Motorcycle	518	0.7
226	Van	430	0.6

vt_fatal = (
    vt[vt["is_fatal"]]
      .groupby("vehicle_type_code_1")
      .agg(fatal_crashes=("collision_id", "count"))
      .reset_index()
      .sort_values("fatal_crashes", ascending=False)
)

total_fatal = vt_fatal["fatal_crashes"].sum()
vt_fatal["pct_of_fatal"] = round(vt_fatal["fatal_crashes"] / total_fatal * 100, 1)

vt_fatal_clean = vt_fatal[vt_fatal["vehicle_type_code_1"] != "Unspecified"]

vt_fatal_clean

	vehicle_type_code_1	fatal_crashes	pct_of_fatal
14	Sedan	45	33.3
15	Station Wagon/Sport Utility Vehicle	33	24.4
12	Motorcycle	20	14.8
16	Taxi	5	3.7
4	Bus	4	3.0
13	Pick-up Truck	3	2.2
3	Box Truck	3	2.2
8	E-Scooter	3	2.2
10	Flat Rack	2	1.5
17	Tractor Truck Diesel	2	1.5
6	Convertible	2	1.5
7	Dump	2	1.5
0	AMBULANCE	1	0.7
11	Motorbike	1	0.7
1	Ambulance	1	0.7
9	FORK LIFT	1	0.7
5	Concrete Mixer	1	0.7
2	Bike	1	0.7
19	Van	1	0.7

vt_merge = vt_all_clean.merge(vt_fatal_clean, on="vehicle_type_code_1", how="left")
vt_merge["fatal_crashes"] = vt_merge["fatal_crashes"].fillna(0)
vt_merge["pct_of_fatal"] = vt_merge["pct_of_fatal"].fillna(0)

vt_merge.head(10)

	vehicle_type_code_1	total_crashes	pct_of_total	fatal_crashes	pct_of_fatal
0	Sedan	34349	45.9	45.0	33.3
1	Station Wagon/Sport Utility Vehicle	27541	36.8	33.0	24.4
2	Taxi	2768	3.7	5.0	3.7
3	Pick-up Truck	1882	2.5	3.0	2.2
4	Box Truck	1417	1.9	3.0	2.2
5	Bus	950	1.3	4.0	3.0
6	Bike	882	1.2	1.0	0.7
7	Tractor Truck Diesel	587	0.8	2.0	1.5
8	Motorcycle	518	0.7	20.0	14.8
9	Van	430	0.6	1.0	0.7

fig = go.Figure()

fig.add_trace(go.Bar(
    x=vt_merge["vehicle_type_code_1"].head(10),
    y=vt_merge["pct_of_total"].head(10),
    name="% of Total Collisions",
    text=vt_merge["pct_of_total"].round(1).astype(str) + "%",
    textposition="outside"
))

fig.add_trace(go.Bar(
    x=vt_merge["vehicle_type_code_1"].head(10),
    y=vt_merge["pct_of_fatal"].head(10),
    name="% of Fatal Collisions",
    text=vt_merge["pct_of_fatal"].round(1).astype(str) + "%",
    textposition="outside"
))

fig.update_layout(
    title="Vehicle Types Involved — Total vs Fatal Collisions",
    xaxis_title="Vehicle Type",
    yaxis_title="Percentage (%)",
    barmode="group",
    template="plotly_white",
    yaxis=dict(range=[0, max(vt_merge["pct_of_total"].max(), vt_merge["pct_of_fatal"].max()) * 1.2]),
)

fig.show()

injured = pd.DataFrame({
    "category": ["Motorists", "Pedestrians", "Cyclists"],
    "count": [
        df["number_of_motorist_injured"].sum(),
        df["number_of_pedestrians_injured"].sum(),
        df["number_of_cyclist_injured"].sum()
    ]
})

fig = px.pie(
    injured,
    names="category",
    values="count",
    title="Distribution of Injured People",
    color_discrete_sequence=px.colors.sequential.Blues
)

fig.show()

killed = pd.DataFrame({
    "category": ["Motorists", "Pedestrians", "Cyclists"],
    "count": [
        df["number_of_motorist_killed"].sum(),
        df["number_of_pedestrians_killed"].sum(),
        df["number_of_cyclist_killed"].sum()
    ]
})

fig = px.pie(
    killed,
    names="category",
    values="count",
    title="Distribution of Fatalities",
    color_discrete_sequence=px.colors.sequential.Reds
)

fig.show()